Gendered language is the use of words that indicate an individual's gender. Though useful in certain context, it can reinforce gender stereotypes and introduce bias, particularly in machine learning models used for tasks like occupation classification. When textual content such as biographies contains gender cues, it can influence model predictions, leading to unfair outcomes such as reduced hiring opportunities for women. To address this issue, we propose GenWriter, an approach that integrates CaseBased Reasoning (CBR) with Large Language Models (LLMs) to rewrite biographies in a way that obfuscates gender while preserving semantic content. We evaluate GenWriter by measuring gender bias in occupation classification before and after rewriting the biographies used for training the occupation classification model. Our results show that GenWriter significantly reduces gender bias by 89% in nurse biographies and 62% in surgeon biographies, while maintaining classification accuracy. In comparison, an LLM-only rewriting approach achieves smaller bias reductions (by 44% and 12% in nurse and surgeon biographies, respectively) and leads to some classification performance degradation.
Iterative design is a well-known methodology that involves prototyping, testing, analyzing, and refining products or processes. Rapid prototyping and evaluation are crucial, enabling designers to quickly identify and resolve issues and iteratively improve the design. However, effective iterative design relies on tools that accelerate both prototype creation and visual evaluation. This paper aims to address this challenge by presenting a framework specifically designed to enhance the iterative design process for CBR applications. In this context, we emphasize the manual and knowledge-intensive task of defining similarity measures. Furthermore, we introduce a proof-of-concept implementation of this framework based on the CBRkit toolkit and the SimViz visualization tool. We studied the capabilities to support the iterative design of CBR applications through a case study in the prototypical cars domain.
Delimiting mobile genetic elements (MGEs) in bacterial genomes traditionally relies on experts in biology. This paper presents a process-oriented case-based reasoning (PO-CBR) approach to formalize and automate expert reasoning for MGE delimitation. In this study, expert knowledge is represented by process cases, where each case follows an ordered stepwise reasoning process based on genomic data on MGEs. These cases form adaptive computational workflows that can be applied to various types of bacterial strains and MGEs. A key feature of this approach is an adaptive mechanism that allows reasoning failures to trigger automatic adaptation rules. Tested on 254 manually annotated MGEs across 124 bacterial genomes, our approach successfully delimited 96.8% of these elements with high accuracy. This study demonstrates the feasibility of encoding biological expertise into a structured automated reasoning system that offers a reliable alternative to identify MGEs in bacterial genomes.
In recent years, the threat and damage caused by ransomware have been steadily increasing. Although many detection methods have been proposed, network security remains a severe challenge due to the continuous emergence of new attack techniques, data imbalance, low detection rates of existing methods, and the lack of interpretability in model decision-making. This study proposes a Case-Based Reasoning with Diffusion Model for Ransomware Detection (CBR-DRD), which classifies network traffic information without feature loss or redundancy by converting it into RGB images. The dataset is then augmented using image generation based on a U-Net diffusion model. Features extracted by a Swin Transformer are used to construct a case base, and a weighted K-Nearest Neighbors (KNN) algorithm is employed to classify the traffic by computing the similarity between a given sample and existing cases, to determine whether it is associated with ransomware activity. Compared with two similar ransomware detection approaches and commonly used generative models, namely the GAN model and a ResNet-based diffusion model designed to address data imbalance, the proposed method achieves superior performance on both the USTC-TFC2016 and ISOT datasets.
To clarify how artificial intelligence models for image classification obtain their predictions, instance-based explanations give examples of other predictions obtained by those models for similar images. XAI (eXplainable Artificial Intelligence) techniques find those examples by getting images similar to the query using similarity metrics that consider image features. In this work, we introduce a Case-Based Reasoning approach that obtains these types of explanations (explanations by examples and counterfactuals) using Visual Question Answering techniques that take advantage of the high performance achieved by Large Language Models. We have evaluated our approach through an offline evaluation, in a context where images to classify and explain may be low quality. We obtained promising results and confirmed the benefits of using multimodal models in XAI, combining knowledge focused on images and natural language.
The time required to retrieve a query's nearest neighbor may quickly become a CBR system's bottleneck when its case base contains a large volume of cases. Approximate retrieval techniques that build and employ complex index data structures can mitigate this issue by providing an acceptable tradeoff between time complexity and retrieval accuracy. In this paper, we propose a specialized multigraph-based index structure composed of multiple nested sub-case bases with individually labeled edge sets. We develop tailored algorithms for both constructing the index and utilizing it in a more efficient retrieval process, and we also evaluate our approach empirically using established benchmark datasets.
The adaptation knowledge container of a CBR system is often represented by adaptation rules that can be learned from the case base using various approaches proposed in the CBR literature. However, the formal representation of such rules has been much less investigated. This paper introduces a formalism for representing such rules, given a formalism that can be used to express cases in an attribute-value formalism. One benefit of this study is a case retrieval algorithm for which the retrieved case is the best one with respect to adaptation (i.e. the one that requires the least adaptation effort as defined by the cost of adaptation rule sequences). Moreover, when attributes have Boolean range, its complexity is independent of the size of the case base.
